IPv6 traffic crosses the 50% mark
68 points by dhruvp
68 points by dhruvp
I still continue to be amused and frustrated that IPv6 adoption goes up at weekends - consumer ISPs and mobile networks are implementing it, business ISPs and corporate IT departments are not.
Anyone know why Greenland and Iceland have 0%?
I'm quite surprised at China being so low, I thought they were one of the fastest IPv6 adopters. Maybe they just mostly don't use Google?
I'm quite surprised that Firefox on Android seems to prefer IPv4, even when both are available. Not sure if this is an Android thing or a Firefox thing.
The Google data is skewed. this one is skewed in different ways: https://stats.labs.apnic.net/ipv6/
The number for China in that one is more plausible, but both seem to agree that Greenland has basically no IPv6. Iceland is a bit higher in that one, but still very low.
France just did a big FTTP push, which included fibre to rural properties. This meant that everyone got a new home router, which probably pushed up their numbers a lot (everyone now has a v6-capable router).
I wonder where the cut-off threshold is. I'd be quite tempted to make some things I self host v6-only if I could be certain that I'd always have v6 available - v6 or VPN is another option. There's complexity from supporting both and at some point v6 is just easier and the benefit of supporting v4 isn't worth it. For a site like Google, that's probably 99.something% adoption. For smaller things, I wonder if it's closer to 60-70%.
The (three or four) ISPs in China are, IIRC, required to implement IPv6 by some administrative regulations. However, a consumer WiFi access point my family bought a few years ago hid had IPv4-only by default, and only had IPv6 opt-in in "Advanced Settings".
There are also some services that my family uses with broken AAAA records, poor routing, etc., such that I had to give them a toggle to globally disable IPv6 on their network. (Also, they implemented DHCPv6 PD incorrectly, and I had to explicitly specify a prefix length in dhcp6c.)
Also, for some reason, many consumer-facing apps don't work well with NAT64, so I've kept all my networks on dual-stack.
I host some personal stuff from home on ipv6 to avoid paying the 5 bucks a month.
My experience has been:
The interesting thing came with trying to host XMPP and Matrix servers. Just absolutely no luck getting anything working there. My impression is that way too much infra in that space just refuses to route on ipv6. Extremely annoying given that I feel like for Matrix servers in particular they must mostly be in "the cloud" or on decent networks.
I guess its also somewhat likely that your websites audience may skew towards having fiber and/or better ISP's and so a higher proportion than the general population will have IPv6.
France just did a big FTTP push, which included fibre to rural properties. This meant that everyone got a new home router, which probably pushed up their numbers a lot (everyone now has a v6-capable router).
... and at the same time, Google French AS don't seem to be pushing hard on IPv6: https://stats.labs.apnic.net/ipv6/FR
Google is blocked in China, so they number you are seeing is the adoption rate of various VPN services.
Google is largely blocked by the Chinese censorship system. To use it you need to route through a VPN, use a proxy, or otherwise obscure your connection so it doesn't get reset.
I've been trying to get IPV6 here in the UK, and only one ISP I've used in the last 20 years has supported it, so I struggle to believe that it is over 50% here. Google reckons 53%, and interestingly apnic says 54% capable (which is an interesting word choice).
BT/EE, Sky, Vodafone, Zen support it on home broadband, so do other smaller ISPs (like A&A, the one I use!). Notably TalkTalk and Virgin Media do not.
EE support it on mobile. Some of the MVNOs and altnets do too.
I'm currently stuck with Virgin Media, and the only provider I've had over the years who supported IPv6 was A&A...
I am looking at a Smarty mobile contract which does support IPv6 as a backup / backhaul connection so I'll need to actually learn some IPv6, only 28 years late ;~)
OpenReach's fibre supports it. The BT Home Hub thing works nicely and IPv6 just works with no custom configuration.
I remember Virgin Media saying 'we have loads of IPv4 addresses so we have no need for IPv6' 15 years ago. And it sounds like their position hasn't changed.
On Openreach based connections it's entirely dependent on what ISP you're using - eg BT/EE support it, Plusnet do not.
It would be so interesting to compare against another world where IPv6 is simply "IPv4 with twice the bits". Would the lack of additional complexity have made the transition easier? Or would the lack of additional features have removed a significant carrot that's driving transition?
Personally, my money would be on the former. I mean every single time I have to enter a v6 address I have to google how to get the software to interpret a v6 address as an IP address rather than a malformed domain or search query. And too often, the answer is "add an entry to /etc/hosts". Unacceptable UX.
It would be so interesting to compare against another world where IPv6 is simply "IPv4 with twice the bits". Would the lack of additional complexity have made the transition easier?
Obviously. People came up with plenty of backwards compatible strategies that would have avoided IPv6. There were reasons that never happened, but they have nothing to do with it not being possible.
Every "backwards compatible" proposal I've read tends to have more or less exactly the same issues is IPv6 adoption hits, because you don't magically make 1 number into 2^64 numbers, right?
Like if your backwards compatible strategy relies on hardware upgrades anyways to do fancy routing are you really backwards compatible? If you end up with "IPv4+-only addresses" then are you backwards compatible? I guess "IPv4 address that only has one port" is kinda backwards compatible but imagine being on the "you're allowed exactly one port" side of things...
I do think there's an argument in "well if we just increase the bit count and change nothing else then people still sit around doing NAT even if they don't need to" and so it ends up being less scary, but at the implementation level you have to do something right? The hardware and software doing the routing still has to be able to handle this new thing, right?
People came up with plenty of backwards compatible strategies that would have avoided IPv6.
The working mechanisms for extending IPv4 are CIDR and NAT. All the schemes I have seen that claim to be able to extend the IPv4 address space in a backwards-compatible manner had obvious flaws that make them just as incompatible as IPv6.
All the schemes I have seen that claim to be able to extend the IPv4 address space in a backwards-compatible manner had obvious flaws that make them just as incompatible as IPv6.
A lot of these schemas are incompatible of the internet today, but not of the internet of when they were introduced. RFC 1710 (SIPP) existed and was a pretty serious proposal for the time.
I'm not sure what obvious flaws you see in SIPP that would have made it as incompatible.
Section 2.2 of RFC 1710 says that the transition mechanism relies on globally-unique IPv4 addresses. There’s a brief overview of transition machinery in section 5 but it doesn’t go into enough detail to describe how it actually works. But the requirement for unique IPv4 addresses is enough of a flaw that further details aren’t necessary.
The transition will be much easier if IPv4 address are still globally unique. The two transition requirements which are the most important are flexibility of deployment and the ability for IPv4 hosts to communicate with IPng hosts. There will be IPng-only hosts, just as there will be IPv4-only hosts. The capability must exist for IPng-only hosts to communicate with IPv4-only hosts globally while IPv4 addresses are globally unique.
This transition plan was already obsolete when it was published in 1994, because IPv4 addresses were no longer globally unique. The precursor to RFC 1918 was published earlier that year, and the Cisco PIX was also released that year. The key selling point of the PIX was NAT.
I can't find it again, but recently there was an blog post attempting to play out this alternate history, and the end result wasn't great. The "extra bits" have to be treated as a postfix to the "regular bits" because the whole point is to be able to do the routing entirely with the regular bits. So you still have a mixture of "small and large" addressing schemes with arbitrary barriers in it, and have to invent similar translation mechanisms to get it to work.
Aside from the extra address space, most of the IPv6 features simplify management rather than adding complexity. It's the IPv4 interop where things get crazy.
Aside from the extra address space, most of the IPv6 features simplify management rather than adding complexity.
The problem is that this fundamentally can not possibly be true, because all the IPv6 features are in addition to IPv4 and have to be supported alongside IPv4, for what seems like at least another 4 decades. Quite possibly forever. All differences become another thing to manage; you can't just assume that everything gets a DHCP-managed address because now all your devices are supposed to self assign some of their addresses and expect DHCP for others, etc.
(Yes I know DHCP6 exists, it's an example. But all b6 evangelists I've met seem to think DHCP6 is an abomination so I don't think "just use DHCP6" would be a solution espoused by its proponents. In any case, it's just an example.)
IPv4 and IPv6 are different things. IPv6 by itself solved a lot of IPv4 management issues. It's the translation mechanisms introduced because you're using both things that make it hard.
Of course in the bigger picture, there is no world where IPv4 disappears, and now we know it was wildly overoptimistic to think that there was. That was the point of the post I can't find: just "expanding IPv4" sounds "simpler" but would still put you in a world where you have two different protocols to translate between. It wouldn't end up making the transition much easier.
There is one Internet. There used to be one Internet protocol 30 years ago. Now there's two, and the existence of IPv6 alongside IPv4 does not solve IPv4 management issues.
Yep, and the existence of "extended IPv4" alongside IPv4 wouldn't either.
Your original question was "Would the lack of additional complexity have made the transition easier?" I think the complexity comes from the transition, not from IPv6 itself, so the answer is no, the transition to "IPv4 with twice the bits" would have been approximately the same.
The design of ipv6 was a mistake.
Care to expand on that?
(I genuinely don't know enough about it (yet) to have an opinion either way, so this is an honest question. I've been ignoring it as "something I don't need to know about yet" but this 50% threshold makes me think that it's probably time I looked into it. lol)